diff options
| author | Shashank Sharma <[email protected]> | 2024-11-20 17:59:49 +0000 |
|---|---|---|
| committer | Alex Deucher <[email protected]> | 2025-04-08 20:48:18 +0000 |
| commit | b0328087c179f47ea6558c3b91b4487c5e10deda (patch) | |
| tree | 7b41fee39432946812413fccfd15364beac238e0 /drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c | |
| parent | drm/amdgpu: add userqueue suspend/resume functions (diff) | |
| download | kernel-b0328087c179f47ea6558c3b91b4487c5e10deda.tar.gz kernel-b0328087c179f47ea6558c3b91b4487c5e10deda.zip | |
drm/amdgpu: suspend gfx userqueues
This patch adds suspend support for gfx userqueues. It typically does
the following:
- adds an enable_signaling function for the eviction fence, so that it
can trigger the userqueue suspend,
- adds a delayed work to handle suspending of the eviction_fence
- adds a suspend function to handle suspending of userqueues which
suspends all the queues under this userq manager and signals the
eviction fence,
- adds a function to replace the old eviction fence with a new one and
attach it to each of the objects,
- adds reference of userq manager in the eviction fence container so
that it can be used in the suspend function.
V2: Addressed Christian's review comments:
- schedule suspend work immediately
V4: Addressed Christian's review comments:
- wait for pending uq fences before starting suspend, added
queue->last_fence for the same
- accommodate ev_fence_mgr into existing code
- some bug fixes and NULL checks
V5: Addressed Christian's review comments (gitlab)
- Wait for eviction fence to get signaled in destroy,
don't signal it
- Wait for eviction fence to get signaled in replace fence,
don't signal it
V6: Addressed Christian's review comments
- Do not destroy the old eviction fence until we have it replaced
- Change the sequence of fence replacement sub-tasks
- reusing the ev_fence delayed work for userqueue suspend as well
(Shashank).
V7: Addressed Christian's review comments
- give evf_mgr as argument (instead of fpriv) to replace_fence()
- save ptr to evf_mgr in ev_fence (instead of uq_mgr)
- modify suspend_all_queues logic to reflect error properly
- remove the garbage drm_exec_lock section in wait_for_signal
- grab the userqueue mutex before starting the wait for fence
- remove the unrelated gobj check from signal_ioctl
V8: Added race condition fixes
Cc: Alex Deucher <[email protected]>
Cc: Christian Koenig <[email protected]>
Acked-by: Christian Koenig <[email protected]>
Signed-off-by: Shashank Sharma <[email protected]>
Signed-off-by: Arvind Yadav <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c | 110 |
1 files changed, 110 insertions, 0 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c index e946c6d1dd6b..43fff8869ac9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c @@ -60,6 +60,16 @@ amdgpu_userqueue_cleanup(struct amdgpu_userq_mgr *uq_mgr, { struct amdgpu_device *adev = uq_mgr->adev; const struct amdgpu_userq_funcs *uq_funcs = adev->userq_funcs[queue->queue_type]; + struct dma_fence *f = queue->last_fence; + int ret; + + if (f && !dma_fence_is_signaled(f)) { + ret = dma_fence_wait_timeout(f, true, msecs_to_jiffies(100)); + if (ret <= 0) { + DRM_ERROR("Timed out waiting for fence f=%p\n", f); + return; + } + } uq_funcs->mqd_destroy(uq_mgr, queue); amdgpu_userq_fence_driver_free(queue); @@ -67,6 +77,22 @@ amdgpu_userqueue_cleanup(struct amdgpu_userq_mgr *uq_mgr, kfree(queue); } +int +amdgpu_userqueue_active(struct amdgpu_userq_mgr *uq_mgr) +{ + struct amdgpu_usermode_queue *queue; + int queue_id; + int ret = 0; + + mutex_lock(&uq_mgr->userq_mutex); + /* Resume all the queues for this process */ + idr_for_each_entry(&uq_mgr->userq_idr, queue, queue_id) + ret += queue->queue_active; + + mutex_unlock(&uq_mgr->userq_mutex); + return ret; +} + #ifdef CONFIG_DRM_AMDGPU_NAVI3X_USERQ static struct amdgpu_usermode_queue * amdgpu_userqueue_find(struct amdgpu_userq_mgr *uq_mgr, int qid) @@ -202,6 +228,7 @@ amdgpu_userqueue_destroy(struct drm_file *filp, int queue_id) amdgpu_bo_unpin(queue->db_obj.obj); amdgpu_bo_unref(&queue->db_obj.obj); amdgpu_userqueue_cleanup(uq_mgr, queue, queue_id); + uq_mgr->num_userqs--; mutex_unlock(&uq_mgr->userq_mutex); return 0; } @@ -277,6 +304,7 @@ amdgpu_userqueue_create(struct drm_file *filp, union drm_amdgpu_userq *args) goto unlock; } args->out.queue_id = qid; + uq_mgr->num_userqs++; unlock: mutex_unlock(&uq_mgr->userq_mutex); @@ -317,11 +345,93 @@ int amdgpu_userq_ioctl(struct drm_device *dev, void *data, } #endif +static int +amdgpu_userqueue_suspend_all(struct amdgpu_userq_mgr *uq_mgr) +{ + struct amdgpu_device *adev = uq_mgr->adev; + const struct amdgpu_userq_funcs *userq_funcs; + struct amdgpu_usermode_queue *queue; + int queue_id; + int ret = 0; + + userq_funcs = adev->userq_funcs[AMDGPU_HW_IP_GFX]; + + /* Try to suspend all the queues in this process ctx */ + idr_for_each_entry(&uq_mgr->userq_idr, queue, queue_id) + ret += userq_funcs->suspend(uq_mgr, queue); + + if (ret) + DRM_ERROR("Couldn't suspend all the queues\n"); + return ret; +} + +static int +amdgpu_userqueue_wait_for_signal(struct amdgpu_userq_mgr *uq_mgr) +{ + struct amdgpu_usermode_queue *queue; + int queue_id, ret; + + idr_for_each_entry(&uq_mgr->userq_idr, queue, queue_id) { + struct dma_fence *f = queue->last_fence; + + if (!f || dma_fence_is_signaled(f)) + continue; + ret = dma_fence_wait_timeout(f, true, msecs_to_jiffies(100)); + if (ret <= 0) { + DRM_ERROR("Timed out waiting for fence f=%p\n", f); + return -ETIMEDOUT; + } + } + + return 0; +} + +void +amdgpu_userqueue_suspend(struct amdgpu_userq_mgr *uq_mgr) +{ + int ret; + struct amdgpu_fpriv *fpriv = uq_mgr_to_fpriv(uq_mgr); + struct amdgpu_eviction_fence_mgr *evf_mgr = &fpriv->evf_mgr; + + mutex_lock(&uq_mgr->userq_mutex); + + /* Wait for any pending userqueue fence to signal */ + ret = amdgpu_userqueue_wait_for_signal(uq_mgr); + if (ret) { + DRM_ERROR("Not suspending userqueue, timeout waiting for work\n"); + goto unlock; + } + + ret = amdgpu_userqueue_suspend_all(uq_mgr); + if (ret) { + DRM_ERROR("Failed to evict userqueue\n"); + goto unlock; + } + + /* Signal current eviction fence */ + amdgpu_eviction_fence_signal(evf_mgr); + + /* Cleanup old eviction fence entry */ + amdgpu_eviction_fence_destroy(evf_mgr); + +unlock: + mutex_unlock(&uq_mgr->userq_mutex); +} + int amdgpu_userq_mgr_init(struct amdgpu_userq_mgr *userq_mgr, struct amdgpu_device *adev) { + struct amdgpu_fpriv *fpriv; + mutex_init(&userq_mgr->userq_mutex); idr_init_base(&userq_mgr->userq_idr, 1); userq_mgr->adev = adev; + userq_mgr->num_userqs = 0; + + fpriv = uq_mgr_to_fpriv(userq_mgr); + if (!fpriv->evf_mgr.ev_fence) { + DRM_ERROR("Eviction fence not initialized yet\n"); + return -EINVAL; + } return 0; } |
